Cornell University Robot Adapts to Injury When It Loses a Limb

Robots are many things but adaptable to their environment usually isn’t one of them. They are rigidly programmed to function within the world around them, and if something goes wrong, there is no “fix it” switch.

This Cornell university robot is not like others of his ilk. His four legs are not pre-programmed to walk. This one works out his own model of himself as he goes along. Just as a newborn would explore its new world, so does this robot, as he learns about his limbs and their relationship to movement. If a leg is damaged, the robot patiently reworks the walking process until he creates a new and effective means of locomotion, which is limping.

A simple four-legged device, researchers believe that this test robot could be the creative seed for more complex robots that can deal with uncertain situations, like space exploration. Ironically, these metal creations may ultimately lend significant insight into both human and animal behavior.

Conducted by a team in the Cornell Computational Synthesis Lab headed by Hod Lipson, assistant professor of mechanical and aerospace engineering, the research, was presented in the journal, Science. The robot’s ability to figure out its own parts and to learn to walk enables it to adapt to different conditions, and find a new gait when it is damaged.

In Lipson’s own words:

“Most robots have a fixed model laboriously designed by human engineers. We showed, for the first time, how the model could emerge within the robot. It makes robots adaptive at a new level, because they can be given a task without requiring a model. It opens the door to a new level of machine cognition and sheds light on the age-old question of machine consciousness, which is all about internal models.”

The robot resembles a four-armed starfish. It begins its animated existence knowing only that it has parts to work with. How these different parts will help in locomotion and their use becomes the application of theory followed by experiment followed by refined theory, or in other words, trial and error.

Lipson also added:

“The machine does not have a single model of itself; it has many, simultaneous, competing, different, candidate models. The models compete over which can best explain the past experiences of the robot.”

A machine that can re-invent itself?

Cool idea, no?

Leave a Comment